[lkml]   [2001]   [Jan]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectUDP performance drop from -test9 to 2.4.0 (fwd)
sorry for the resend. i sent this earlier today but still haven't seen it,
so i'm resending without attachments.

the originally attached netperf results are at:

---------- Forwarded message ----------
Date: Mon, 8 Jan 2001 09:38:24 -0700 (MST)
From: Todd <>
Subject: UDP performance drop from -test9 to 2.4.0


i'm seeing a signficant performance decline between the -test9 and the
release version 2.4.0 on udp on acenic gig-e cards. i should say that
the performance is startlingly better than it was under 2.2.14 (where the
best we could get was 150+ Mbps on udp) but it still represents a
signficant decline.

i'm attaching netperf numbers with 1500 and 9000 byte mtus for both kernel
versions (with interrupt coalescing and everything else as default--this
is just a compile/boot/test scenario with no tuning in particular).
standard kernel, standard /proc settings, standard acenic firmware and
driver that shipped with the version in question.

the summary of results: 12% reduction in performance at 1500-byte mtu, 6%
reduction at 9000. when looking at the results remember that the second
line in each case is the 'receive bandwith' line, which is the only useful
number to report.

kernel version max udp bandwith mtu message size
============== ================ === ============
2.4.0-test9 554.9 1500 1470
2.4.0-test9 636.7 9000 8870
2.4.0 489.4 1500 1470
2.4.0 597.3 9000 7270

both versions show the horrible effect that the reassembly of fragments
has on receive bandwith, but that's not really my question for the day

any info would be helpful. i'd be happy to run any further tests to get
at the bottom of this.


Todd Underwood,

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
Please read the FAQ at

 \ /
  Last update: 2005-03-22 12:52    [W:0.040 / U:1.136 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site